A Novel Efficient Density-Based Clustering Algorithm Using Cross and Rotating Expansion Technique

• Main Authors: She, Po-Yi, 佘柏毅
• Other Authors: Tsai, Cheng-Fa
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/58781996689408058679

碩士 === 國立屏東科技大學 === 資訊管理系所 === 101 === This thesis develops a novel clustering technique called SPY_DBSACN. It presents a new density-based clustering technique with diagonal sampling and a new way of fold and rotation. The SPY_DBSCAN’s expansion without selecting data points to increase computation cost and it may considerably reduce time cost. The experimental results confirm that the proposed technique has very high clustering accuracy and noise filtering rate, and is faster than the well-known density-based data clustering approaches such as DBSCAN, IDBSCAN and KIDBSCAN schemes. The thesis has the merits as follows: 1) the proposed method utilizes cross method to form similarly rotate expansion of clustering, 2) the proposed approach can decrease the data points in the seeds so it can reduce the time of processing. According to the simulation results, the SPY_DBSCAN may lower about sixty-seven percent time cost for performing data clustering comparing with IDBSCAN in 100,000 points dataset experiment. It even decreases more than eighty-six percent time cost for conducting data clustering comparing with DBSCAN in more than 100,000 points dataset experiment, 3) the SPY_DBSCAN has very high clustering accuracy and noise filtering rate, both close to one hundred percent, 4) the proposed approach is very simple and easily to apply in numerous related fields.